A pathologist's assessment is considered the gold standard for deciding whether cancer is present or absent and whether atypical epithelium is reported. However, extensive variation between pathologists has been documented, with disagreement in diagnosis of atypical ductal hyperplasia (ADH) in more than 50% of cases and a misclassification rate for cases designated by experts as ductal carcinoma in situ (DCIS) of 17%. We propose to evaluate the extent of variability, its sources and its impact on women undergoing breast biopsy. Our research will also evaluate possible techniques to improve accuracy, such as double reading. To accomplish our study aims we will develop four diagnostic test sets, weighted to include a high percentage of cases with ADH and DCIS, as defined by a panel of international experts. More than 100 U.S. pathologists will independently review the test sets during two phases, separated by 6 months. Our specific aims are: 1. To assess the accuracy of pathologists in the interpretation of breast pathology specimens and quantify the extent and possible sources of intra- and inter-observer variability. 2. To assess whether independent double reading by two pathologists can improve the accuracy of interpretation of breast pathology specimens. 3. To quantify the implications of variability among pathologists' interpretations of breast specimens on short-term patient care and associated resource utilization within the U.S. population. We hypothesize that the following characteristics are associated with lower accuracy: a) Patient characteristics (age < 50 years, dense breast, extensive background proliferative change, ADH and DCIS interpretation): and b) Pathologist characteristics (e.g., less clinical experience). In summary, we will evaluate the accuracy of breast pathology interpretation, emphasizing the classification of ADH and DCIS, where previous studies reveal a concerning degree of diagnostic errors. This large multi- center study of community pathologists addresses a topic that has growing clinical importance. The proposed work is innovative in that we will go beyond simply quantifying the existence of errors. We will study patient and physician characteristics associated with inaccurate diagnoses. We will then evaluate methods for improving health care delivery using double reading. Finally, we will use a systematic method to make projections from our data to clinical implications of inaccurate diagnoses.